This repository contains the implementation for the paper Generating High Dimensional User-Specific Wireless Channels using Diffusion Models (https://www.arxiv.org/abs/2409.03924).

After installing QuaDRiGa (https://quadriga-channel-model.de/software/), place main_chgen.m in the /quadriga_src/ folder.

Then, execute the file in MATLAB. After generation, place the output files into /data/QuaDRiGa. Alternatively, you can download the dataset from Google Drive here.

To create and activate the Conda environment using the provided environment.yml, follow these steps:

1. Create the environment:

   conda env create -f environment.yml

2. Activate the environment:

   conda activate cDDIM

First, create /cDDIM_10000/ folder, and execute script_channel_ddim.py to train the model:

python script_channel_ddim.py

For inference, use the following commands:

python ddim_inference.py generate

to generate channel matrices. Then,

python ddim_inference.py concatenate

to concatenate the generated matrices. The above description is for the quadriga dataset. A version for the DeepMIMO dataset will be updated.

This repository was inspired by the following codebases:

• The codebase is primarily based on conditional MNIST: https://github.com/cloneofsimo/minDiffusion

Two downstream tasks mentioned in the paper:

• Channel compression - CRNet: https://github.com/Kylin9511/CRNet
• Site-specific beamforming - DLGF: https://github.com/YuqiangHeng/DLGF

Other ideas are referenced in the paper.

If you find this repository helpful, please cite our work!:

@article{lee2024generating,
  title={Generating High Dimensional User-Specific Wireless Channels using Diffusion Models},
  author={Lee, Taekyun and Park, Juseong and Kim, Hyeji and Andrews, Jeffrey G},
  journal={arXiv preprint arXiv:2409.03924},
  year={2024}
}